Google on Thursday released Gemini 4 Hydrogen, a fast new member of its Gemini 4 family, and made it the default model in AI Mode, the chatbot-style part of Google Search, for users worldwide.
“Hydrogen is now available globally in Google Search for all users,” Robby Stein, Google’s vice president of product for Search, wrote in a company blog post, calling it “the biggest model upgrade to Search yet.” The Verge reported that Stein said Hydrogen is now the default model in AI Mode and is being used for the hardest questions in AI Overviews.
Google positions Hydrogen below its two frontier models, Gemini 4 Carbon and Gemini 4 Argon, which it says are built for deep, long-horizon reasoning. “Hydrogen is built for rapid execution across diverse agentic workflows,” Koray Kavukcuoglu, Google’s chief AI architect, wrote in the launch post. Google’s developer documentation calls it the recommended upgrade from Gemini 3.8 Flash, and the company has dropped the Flash name for this generation. Ars Technica described Hydrogen as replacing the faster, more economical Flash models most people in Google’s ecosystem use.
Who gets it
Beyond Search, Hydrogen is available on both the free and paid tiers of Google AI Studio and the Gemini API, according to Google’s developer documentation. The free tier is rate-limited. On Google Cloud, the model is generally available in global, U.S. and EU regions.
In the Gemini app, paid subscribers can pick Hydrogen from the model menu. Free users get it only through automatic model selection, “when it’s the appropriate model for the prompt,” according to a footnote in Google’s announcement. Ars Technica reported that free and AI Plus users of the app were still on an older Flash-Lite model, so what a free user actually sees may vary by prompt. The app’s model menu still lists 3.5 Flash-Lite for quick answers, according to 9to5Google.
Hydrogen is also coming to Gemini in Google Sheets for AI Pro and Ultra subscribers and eligible business plans, and to Google’s Antigravity coding tool for AI Pro and Ultra subscribers.
What developers will pay
Google lists Hydrogen at $1.50 per million input tokens and $7.50 per million output tokens on its standard paid tier, with thinking tokens billed as output. Cached input drops to $0.075 per million tokens, and batch and flex processing cost half the standard rate.
That is cheaper than Carbon and Argon, which cost $2 and $10 per million tokens through Nov. 30 and rise to $4 and $20 on Dec. 1, according to Google’s pricing page. But it is not cheaper than the model it replaces. Gemini 3.8 Flash is on a promotional rate of $0.75 and $3.75 through Dec. 31, so Hydrogen costs developers twice as much until Flash’s promotion ends and its price rises to match Hydrogen’s on Jan. 1. Google lists no introductory discount for Hydrogen.
What “agentic” means here
Google says Hydrogen is designed to act, not just answer. Its API supports function calling, code execution, grounding in Google Search and Maps, file search and computer use, and it has a context window of about 1 million tokens. Google says the model runs file reads and searches in parallel instead of one at a time, recovers when a tool call fails, and flags missing context instead of guessing, which the company says cuts fabricated claims by more than five times compared with 3.8 Flash.
In Antigravity, a new Orchestrator feature for Ultra subscribers lets Carbon plan a task while Hydrogen writes edits, updates packages and runs tests, handing back to Carbon when it gets stuck.
How it scores
Google says Hydrogen finishes tasks 2.5 times faster than 3.8 Flash at its default setting and more than triples Flash’s Terminal-Bench 4.1 score, to 57% from 18.2%. The independent benchmarking firm Artificial Analysis lists Hydrogen at 54 on its Intelligence Index at the model’s highest thinking setting, matching Google’s figure. Google says it scores 51 at the default setting, one point ahead of Argon.
Google also says Hydrogen matches or beats competing models in its class, including Anthropic’s Claude Sonnet 5.5 and OpenAI’s GPT-6.1 Terra. Those comparisons come from Google’s own model card and have not been independently verified. The same table shows Hydrogen trailing Claude Sonnet 5.5 on three tests: Terminal-Bench 4.1, OSWorld-2.1 and Terminal-Bench-Science. GPT-6.1 Terra also beats it on Terminal-Bench-Science in Google’s table, 58.3% to 51.4%.
Google said a full safety and security report for Hydrogen will be published soon. Its model card says the model can still hallucinate, may time out, and can use more tokens at higher thinking levels.